Martin, thanks for your feedback.
I am no ASIO expert, but a few things came to mind:
1) As far as I know, it is not safe to do a async_send operation on a socket before the previous async_send operation has completed. I have seen this mentioned somewhere in the documentation, but can't find the place now.
it might be. then again, all send events seem to complete find, as the send callback is called the proper number of times.
2) Line 56: The logic in handle_send_to seems flawed. It issues a async_receive_from()? I would expect it to either send another UDP packet or just check the error code from the original send operation.
this is a leftover from the boost asio sample code I copied it over from, see here: http://www.boost.org/doc/libs/1_44_0/doc/html/boost_asio/example/echo/async_... removing it actually solves my issue. wow - I wouldn't have thought :) now I'm hitting the following barriers: b sent 200000 messages a received 199597 messages b sent 2000000 messages a received 1990728 messages these both seem to be over 99% fidelity, which is, well, sort of OK :)
3) Line 116: The lifetime of the buffer passed to send() may end before the async operation completes. This will likely not cause the issues seen here, but worth mentioning.
actually an asio::buffer object is created in send(), which copies the contents over. this is the same pattern as used in the same sample code quoted above.
Have you tried - just for debugging - changing the send() operation to be synchronous?
actually haven't. Akos